Matrix-Stripe-Cache-Based Contiguity Transform for Fragmented Writes in RAID-5
IEEE Transactions on Computers
Pipelining network storage i/o
ICCS'06 Proceedings of the 6th international conference on Computational Science - Volume Part I
Hi-index | 0.01 |
Linux clusters of commodity computer systems and interconnectshave become the fastest growing choice for buildingcost-effective high-performance parallel computing systems.The Parallel Virtual File System (PVFS) could potentially fulfillthe requirements of large I/O-intensive parallel applications.It provides a high-performance parallel file system bystriping file data across multiple cluster nodes, called I/Onodes. Therefore, the choice of storage devices on I/O nodes iscrucial to PVFS.In this paper, we study the impact of software RAIDs andhardware RAIDs on the performance of PVFS when they areused on I/O nodes. We first establish a baseline performanceof both RAIDs in a stand-alone configuration. We then presentthe performance of PVFS for a workload comprising concurrentreads and writes using ROMIO MPI-IO, and for the BTIObenchmark with a noncontiguous access pattern. We foundthat software RAIDs have a comparable performance to hardwareRAIDs, except for write operations that require file synchronization.